The bottleneck in AI infrastructure expansion is shifting from chip computing power to something more fundamental and harder to quickly address—electricity and labor. At the G20 technology ministers’ meeting held on September 1 in Chapel Hill, North Carolina, Elon Musk and Mark Zuckerberg made a rare joint appearance, sending the same signal to policymakers from around the world: skilled labor shortages and insufficient power supply are becoming the two core obstacles constraining AI development.

Musk placed the power issue at the top of the urgency list. Citing analyst consensus estimates, he said a clear shortage could emerge as early as next year, and by 2027, the power shortfall facing AI chips will reach at least 15 gigawatts.

The structural contradiction behind this judgment lies in a severe growth-rate imbalance: AI chip production capacity is growing at roughly 40% to 50% per year, while available electricity supply outside China is growing at only about 10% to 20% annually. Musk summed up this unsustainable relationship in one sentence: “Obviously, the thing that grows faster will eventually overwhelm the thing that grows slower.”

Energy self-sufficiency becomes a new dimension of AI competition

Musk also revealed a noteworthy development at the meeting: companies such as Google and Anthropic have begun leasing computing capacity from SpaceX, precisely because SpaceX solved its power supply problem first by building its own generation plants, enabling rapid scaling. He said his own companies are building the necessary power generation facilities alongside their data centers.

This detail reveals a deeper shift in the competitive landscape of AI infrastructure. Over the past two years, market attention has focused on GPU supply and data center site selection; now, the ability to secure stable power is becoming a new differentiator. Companies with stronger energy self-sufficiency will enjoy a significant advantage in the pace of compute expansion.

Zuckerberg, speaking via video link, extended the conversation from power to another equally thorny bottleneck—labor. He said the demand for AI data center construction is enormous, potentially requiring “hundreds of thousands or even millions” of skilled technical jobs in the future. He stated bluntly that Meta is already struggling to find enough skilled workers to build data centers—a challenge that is not a distant risk but a constraint happening right now.

The statements from the two tech leaders together paint a clear picture: the pace of AI development is outstripping the carrying capacity of existing infrastructure systems. A shortage of electricity on the hardware side and a shortage of people on the construction and operations side are compounding to significantly lengthen the cycle from data center planning to commissioning.

Demis Hassabis, CEO of Google’s DeepMind, also participated in the discussion via video. He likened AI’s potential impact to “10 times the Industrial Revolution,” while emphasizing that countries must drive development in a responsible manner.

Regulatory divergence and domestic resistance coexist

Beyond infrastructure issues, differences in the regulatory environment became another flashpoint at the meeting. Musk criticized Europe for over-regulating technology, arguing that innovation should develop in a relatively permissive regulatory environment. He advocated that new technologies should be “legal by default” rather than “illegal by default,” and said that while European regulation cannot stop technological progress, it will clearly slow the pace of innovation.

This stance echoes the policy tone in the United States. Trump had criticized opponents of AI data center construction the previous day as “backward and poor,” reflecting the overall inclination of U.S. political circles to accelerate AI infrastructure buildout.

Yet the aggressive push by tech giants is not without resistance even within the United States. Polls show a majority of Americans oppose building more data centers in their local areas, with concerns centered on rising electricity bills and noise pollution. This creates a notable policy paradox: solving the power shortage requires large-scale construction of generation facilities and data centers, but the construction itself is encountering public opposition.

Long-term economic expectations remain optimistic

Despite the many bottlenecks, tech leaders remain highly optimistic about AI’s long-term economic value. Musk estimated that AI could expand the global economy by 20% to 30%, equivalent to adding roughly $20 trillion to $30 trillion per year. He acknowledged this forecast is a “rough estimate,” but said it is one he would “bet heavily on.”

Hassabis framed the potential impact in even broader terms, saying that once human-level AI is achieved, its impact could be 10 times that of the Industrial Revolution.

The following are key figures disclosed by various parties at the meeting:

MetricFigureAI chip power shortfall by 2027At least 15 GWAnnual AI chip capacity growth40% to 50%Annual power supply growth outside China10% to 20%Potential annual global economic gain from AIapproximately $20 trillion to $30 trillionPotential expansion of global economy from AI20% to 30%

Note: The above figures come from Musk’s remarks at the G20 technology ministers’ meeting and analyst consensus estimates.

From a market perspective, the implications of these statements are direct and far-reaching. Competition in AI infrastructure investment has extended from chips and data centers into power generation and energy infrastructure. For investors, power equipment, grid upgrades, energy storage, and engineering services related to data center construction may become the next focal points of capital expenditure. At the same time, the shortage of skilled technical workers means that engineering contractors with the capacity to deliver large-scale projects will gain stronger bargaining power.

Musk and Zuckerberg’s joint appearance at the G20 is, in essence, a message to policymakers: the expansion of the AI industry has outpaced the carrying capacity of public infrastructure. If power supply and talent development cannot keep pace, the vision of AI-driven economic growth will face real constraints.